[debian-mysql] Help with MariaDB 10.5 cross-building?

[Daniel Black]

On Tue, Feb 21, 2023 at 5:54 AM Otto Kekäläinen <otto at debian.org> wrote:
>
> On Mon, 20 Feb 2023 at 13:58, Helmut Grohne <helmut at subdivi.de> wrote:
> > please avoid mailing me on cross compilation questions directly.
> > Instead, please mail debian-cross at lists.debian.org. This reaches a wider
> > audience and increases your chances for a timely response. Thanks.
>
> Noted. You might want to mention this in the footer at
> http://crossqa.debian.net/
>
> ...
> > We can see that mariadb already has a native build pass and exports
> > tools to be used in the cross build. I suppose, uca-dump was added
> > rather recently and was missed here. The big question though is whether
> > uca-dump actually behaves the same on all architectures. Would it be
> > possible for you to run it on various systems and compare the resulting
> > header? Usually the important comparisons are big endian vs little
> > endian and 32bit vs 64bit, which amounts to 3 builds if done right.
> > Another approach is dumping the resulting header to the build log and
> > uploading a thus modified package to experimental. Then one can scrape
> > the header from the build logs.
> >
> > If one assumes that uca-dump behaves the same on different
> > architectures, then fixing this issue is a relatively simple matter of
> > adding it to exported executables. I'm attaching a patch for your
> > convenience. Please only apply it after verifying that its output is
> > indeed architecture-independent.
>
> Thanks Helmut for the advice!

Thanks Helmut for the patch. Looks good to me.

> If one of our contributors (Faustin? Tuukka?) can file an MR at
> https://salsa.debian.org/mariadb-team/mariadb-server/-/merge_requests
> with suitable code for dumping the header as Helmut suggested, I can
> upload it to Launchpad (which has both 64 and 32 bit builders and both
> endianness) so we can see the result in the logs, and if needed I
> could also include that in the next Debian experimental upload.

No real need. uca-dump generates C code from a parsed text file, so
its all is arch independent.

I believe the test main.ctype_utf8mb4_uca_allkeys1400 show this is correct.

Otto, would you like to do a 10.10 PR with the patch?